From AI experiments to organizational shift: Lessons from Mercari’s transformation (Michael Galloway and Snehal Shinde)

Key takeaways

AI adoption must be measurable and linked to organizational value to justify investment and guide efforts.
Building organizational visibility into AI workflows and impacts is crucial for understanding bottlenecks and opportunities.
Cultural change and managing organizational complexity are as important as technological adoption when transforming at scale.
Iterative, honest assessment, and readiness to pivot are vital strategies—embodying the Stockdale Paradox—particularly in complex AI initiatives.
The future of software development involves human-AI collaboration in dynamic loops, emphasizing early-stage AI integration and continuous feedback.

Summary

Mercari’s Scale and Engineering Culture

This section discusses Mercari's vast scale, serving over 23 million monthly active users, and its engineering culture that emphasizes data-driven decision making. It highlights how visibility through data is central to their operations, enabling continuous improvement and product excellence.

Transition to an AI-Native Organization

Mercari's leadership mandated 100% AI adoption across all roles, prompting the organization to rethink workflows and culture. A dedicated AI task force and AI champions from various departments were established to integrate AI deeply into work processes, reflecting a cultural shift towards AI augmentation rather than mere experimentation.

Building Visibility and Measuring AI Impact

The podcast details how Mercari built new dashboards to measure AI adoption, workflow efficiencies, and organizational bottlenecks. They developed systems to stitch data from various tools like GitHub, Datadog, and their AI tools, providing insights into AI's actual impact on productivity, which was previously unquantifiable with existing dashboards.

Lessons and Challenges in AI Adoption

The speakers share key lessons, including understanding the complexity of existing systems, managing organizational change, and the importance of visibility. They also discuss challenges such as system fragility, organizational inertia, and the need for continuous iteration and honest assessment, embodying the Stockdale Paradox—faith in the vision coupled with brutal honesty about current realities.

Future Directions and Innovations in AI-Driven Development

Mercari is exploring agent-based development, refactoring older systems, and refining the SDLC with AI-driven loops. Initiatives like agent spec-driven development aim to bring AI to the front of the development process, reducing human approval steps and enabling faster, safer deployment cycles.

Show notes

Michael Galloway leads Platform Engineering at Mercari, while Snehal Shinde leads Cost and Performance Engineering. Together, they have been at the center of Mercari's effort to become an AI-native company.

In this session from DX Annual, Michael and Snehal share what happened after Mercari's CEO mandated 100% AI adoption across the organization. While AI accelerated code generation and increased engineering output, the team quickly discovered that their existing dashboards could not answer a simple question: was AI actually improving productivity?

They discuss how Mercari built new visibility into AI usage and software delivery, the bottlenecks they uncovered across the SDLC, why faster coding did not automatically translate into faster delivery, and the lessons they learned rolling out AI at scale. They also share how Mercari is rethinking software development around agents, feedback loops, and new ways of working.

In this episode, we cover:

(00:00) Intro

(01:46) Mercari’s scale and engineering culture

(02:51) DX awards at Mercari

(03:44) Mercari’s push to become AI-native

(06:34) The mandate to rethink everything

(08:02) Mercari’s AI visibility problem and how they solved it

(11:30) Mercari’s early findings on AI implementation

(18:47) Closing the AI awareness gap at Mercari

(21:11) Mapping AI opportunities across Mercari

(31:32) Unpacking the results from the second rollout

(34:14) Agent spec-driven development and what’s next